Skip to content
assignmentAssessment and Measurement

Procedural Integrity (Implementation Fidelity)

How closely a procedure was carried out as written — and why that has to be measured, not assumed.

Translated from the Portuguese original. Read the original (Português)

boltIn short

An intervention plan that looks good on paper is worthless if it gets applied halfway in practice. Procedural integrity — also called implementation fidelity or treatment integrity — is how closely a procedure was carried out exactly as planned: no steps skipped, no improvising added. It is measured the same way anything else is measured in Behavior Analysis: steps written in observable terms, a check of what actually happened, and a number that comes out of that count. Without that number, when a person's data do not improve, no one can say whether the procedure does not work or whether it was never actually applied — and the two conclusions point in opposite directions. The good news is that integrity is a behavior of the person applying it: like any behavior, it improves with training, feedback, and workable plans — and it quietly gets worse on its own when no one is watching.

Technical definition

Procedural integrity (procedural fidelity), also called treatment integrity or implementation fidelity, is the degree to which the independent variable — the procedure, intervention, or teaching plan — is carried out exactly as planned and described, without unplanned variables being inadvertently introduced alongside it (Cooper, Heron & Heward, 2020). This requirement applies to every condition of a program or a study, including the baseline and the comparison conditions: they too must happen as described, or else the contrast between conditions stops meaning what it is assumed to mean. In practice, integrity is expressed as a percentage of components (or opportunities) implemented correctly, ideally gathered through direct observation and reported component by component, not just as a single number.

The question only integrity can answer

Picture the graph of a child whose data have not budged in three weeks. The plan called for reinforcing requests made with words and withholding attention from the problem behavior. Faced with flat data, there are two completely different explanations — and they call for opposite decisions. Either the chosen procedure genuinely does not work for this person, and it is time to revisit the function of the behavior and change course. Or the procedure never actually happened the way it is written, and switching it now would mean abandoning something that was never truly tested.

Without a measure of integrity, there is no way to choose between the two. And the cost of getting it wrong is high either way: discarding an effective intervention because it was poorly applied, or sticking with something useless because the assumption is that "the problem must be the application." That is why, in data-based decision making, integrity is the first thing to check when progress stalls — before touching the plan.

The data have not improved. So what should change now? HIGH integrity the plan really was applied The problem is the procedure • it did not work for this person • revisit the function of the behavior • adjust or replace the intervention LOW integrity the plan came out different from the page Nothing to conclude about the plan • train and give feedback to whoever applies it • simplify whatever is not workable • measure again before deciding Without the integrity measure, the two conclusions below are indistinguishable.
The same flat graph allows for two opposite readings. What decides between them is not the intuition of whoever is following the case — it is the integrity data.

There is also a second, less obvious reason. When a procedure is said to be evidence-based, the evidence exists for that procedure, the way it was tested. A shortened version, improvised on the fly, or applied only half the time does not automatically inherit that evidence: strictly speaking, it is a different procedure, one that still has no research behind it. Integrity is the bridge between what research has shown to work and what the person actually received.

The double standard the field took decades to face

Behavior Analysis has always been obsessive about measuring the behavior of the person being served — the dependent variable. The behavior is given an operational definition, a second observer is brought in to record in parallel, and interobserver agreement is calculated. On the other side of the equation, though, for a long time the procedure was merely described — and assumed to have happened.

That asymmetry was called out by Peterson, Homer, and Wonderlich in 1982, in a Journal of Applied Behavior Analysis article whose survey showed that the majority of published studies included no assessment at all of whether the independent variable had actually occurred — and that a far-from-negligible share did not even define it operationally. In other words: the effect was measured rigorously, while the cause was simply taken for granted.

Twenty-five years later, the picture had changed less than one might expect. A review by McIntyre, Gresham, DiGennaro, and Reed (2007) examined 152 school-based intervention studies published in JABA between 1991 and 2005: practically all of them (95%) defined the intervention operationally — but only 30% reported integrity data, and about 45% were judged at high risk of imprecise implementation. Compared with previous decades, progress in reporting integrity was modest.

The topic is still very much alive. In 2023, a journal in the field devoted an entire special issue to procedural fidelity (St. Peter, 2023) — a sign that the question "was this really applied as written?" still does not have an automatic answer.

Three ways a plan can come off the page wrong

"Low integrity" rarely means someone ignored the plan. It is almost always small, well-intentioned deviations that go unnoticed day to day. They fall into three patterns — and they do not carry equal weight:

Failing to do what the plan calls for

This is the failure of absence: not waiting the agreed number of seconds before helping, forgetting to praise a correct response, skipping the planned prompt, not recording the trial, cutting the session short. Each isolated omission looks harmless — which is exactly why they pile up without anyone noticing.

The typical effect is to weaken the procedure: the planned consequence starts occurring inconsistently, learning slows down, and progress stalls below what would otherwise be possible.

Doing what the plan does not call for

This is the failure of addition: giving attention to the problem behavior "just this once," giving in to a request made the wrong way, delivering the reinforcer after an incorrect response, adding an extra prompt that rushes the response. These are the easiest errors to make under pressure — and the most costly.

In a translational study by St. Peter Pipkin, Vollmer, and Sloman (2010), conditions in which the problem behavior was occasionally reinforced were more harmful than simply failing to reinforce the alternative response (the authors also noted that the order in which conditions were presented influenced the outcome as well). The reason is well known: giving in every now and then turns what should be extinction into intermittent reinforcement — precisely the schedule that makes a behavior more resistant to stopping.

Silent erosion over time

Procedural drift is not an error, it is a trend: implementation starts out faithful, right after training, and gradually moves away from the plan week after week — without anyone ever deciding to change anything. A step that was a hassle disappears; a shortcut that "works just as well" takes its place; the prompt comes a little earlier with each session.

It is the strongest argument for measuring integrity routinely, not just at the start. The good news is that drift responds well to a cheap intervention: periodic feedback to the person applying the procedure about their own performance.

How it is measured, in practice

Measuring integrity does not require equipment or a fancy form. It requires the same thing as measuring any behavior: defining what counts as "done" before looking. There are five steps.

  • Write the procedure as observable steps. "Being consistent" is not a step; "wait 5 seconds before giving the prompt" is. The same yardstick as an operational definition applies: two people watching the same scene have to score it the same way.
  • Choose the unit of analysis. It can be by component (of the plan's 10 steps, how many occurred during the session?) or by opportunity (of today's 20 trials, in how many was the step carried out correctly?). The second is usually more granular and helps spot drift within the session itself.
  • Do the math. Correct steps ÷ planned steps × 100.
  • Choose the data source. Direct observation — live or on video — is the most reliable. Permanent products (completed data sheets, handouts delivered, materials used) are low-cost and work well as a complement. Self-report is the most practical and the least accurate: the tendency is to overestimate one's own fidelity, not out of dishonesty, but because no one sees their own automatic habits.
  • Sample well. Varied sessions, different days and times, everyone who applies the plan — including family members, if they apply it at home. A single observation, scheduled ahead of time, mostly measures the behavior of someone who knows they are being watched.
Planned steps of the procedure Did it happen? 1 Prepare the materials before starting 2 Give the instruction once, calmly 3 Wait 5 seconds for the response 4 Praise and deliver the item within 3 seconds 5 Use the agreed-upon prompt when there is an error 6 Withhold attention from the problem behavior 7 Record every trial on the data sheet 8 End with a preferred activity 6 correct steps ÷ 8 planned × 100 = 75%
Integrity is a simple calculation over observable steps. Here, 6 of the 8 steps happened as planned — 75%. But one of the deviations is a commission error (step 6, giving attention to the problem behavior), and that type usually carries more weight than the percentage suggests.
warning

Beware of the single numberAn "85% integrity" figure can hide the fact that the 15% that fail are always the same step — and precisely the decisive step in the procedure. That is why the recommendation is to report by component: one column per step, with the percentage for each. For the same reason, there is no universal cutoff ("90% is good"): how much a failure costs depends on which step failed and how sensitive the procedure is to that particular deviation.

What exactly counts as "implemented"

Outside Behavior Analysis, the same concern shows up in research on prevention and health programs, where Dane and Schneider (1998) organized fidelity into five aspects. They help reveal what a step-by-step checklist, on its own, tends to leave out. Tap the cards:

How to keep integrity high

When integrity drops, the useful question is not "who made the mistake?" but "what, in the environment of the person applying the plan, made the error likely?" The answer is usually found in one of these four areas:

  • Performance-based training. Reading the plan does not teach you how to carry it out. The package that works combines instruction, modeling (watching someone do it), rehearsal (doing it in front of someone), and immediate feedback — repeated until a mastery criterion is met, preferably in the real context where it will be applied.
  • Ongoing performance feedback. This is the strategy with the most consistent empirical support for sustaining integrity after initial training: regularly showing the person applying the plan how their own performance has been. The meta-analysis by Solomon, Klein, and Politylo (2012), on the single-case literature in schools, found a moderate effect — and, more importantly, feedback counteracts the downward trend that shows up when initial training is the only thing that happens.
  • Workable plans. A procedure with 22 steps, requiring two people the school does not have, does not fail because of carelessness: it fails by design. Simplifying it, fitting it into the real routine, and lowering the response cost for whoever applies it raises integrity more than any amount of pressure. This is also part of social validity: a plan no one can sustain is not acceptable.
  • Routine measurement. Integrity measured once is a snapshot; measured continuously, it is an early-warning system. It belongs on the same panel where the progress curve is already being watched — and it is the first line to check before changing anything in the plan.
lightbulb

Measuring is not policingThe integrity figure describes the procedure, not the worth of the person applying it — and whoever applies it is almost always the person with the least decision-making power over the plan: the intern, the teacher with 28 students, the parent after a day of work. Using the number to shame people tends to produce two effects: the records improve and the practice does not. Used as input for supervision — with specific, descriptive, frequent feedback — that same number becomes raw material for training, not a tool for pressure. It is no accident that the BACB's ethics code lists exactly this among a supervisor's responsibilities: tracking supervisees' performance with data and providing documented feedback.

procedural integrity implementation fidelity omission error commission error procedural drift performance feedback

Test your understanding

quiz

Quick quiz

Three questions to lock in the concept. You will see the explanation right away.

Further reading

  1. Cooper, J. O., Heron, T. E. & Heward, W. L. Applied Behavior Analysis. 3rd ed. Hoboken: Pearson, 2020.
  2. Peterson, L., Homer, A. L. & Wonderlich, S. A. "The integrity of independent variables in behavior analysis". Journal of Applied Behavior Analysis, 15(4), 477–492, 1982.
  3. McIntyre, L. L., Gresham, F. M., DiGennaro, F. D. & Reed, D. D. "Treatment integrity of school-based interventions with children in the Journal of Applied Behavior Analysis 1991–2005". Journal of Applied Behavior Analysis, 40(4), 659–672, 2007.
  4. St. Peter Pipkin, C., Vollmer, T. R. & Sloman, K. N. "Effects of treatment integrity failures during differential reinforcement of alternative behavior: a translational model". Journal of Applied Behavior Analysis, 43(1), 47–70, 2010.
  5. Dane, A. V. & Schneider, B. H. "Program integrity in primary and early secondary prevention: are implementation effects out of control?". Clinical Psychology Review, 18(1), 23–45, 1998.
  6. Solomon, B. G., Klein, S. A. & Politylo, B. C. "The effect of performance feedback on teachers' treatment integrity: a meta-analysis of the single-case literature". School Psychology Review, 41(2), 160–175, 2012.
  7. St. Peter, C. C. "On the importance of fidelity, integrity, and adherence in behavior-analytic publications: an introduction to the special issue on procedural fidelity". Education and Treatment of Children, 46(3), 195–199, 2023.
  8. Behavior Analyst Certification Board (BACB). Ethics Code for Behavior Analysts. Littleton, CO: BACB, 2020 (in effect since 2022).
info

This entry is educational and does not replace assessment, diagnosis or treatment by a qualified health professional. Every person is unique — for a specific case, seek individual guidance.

Looking for professional guidance?

Instituto Walden4 is a psychology and education center in Brasília, Brazil, offering evidence-based assessment and care for autism, psychotherapy and development — in person and online. Our team serves in Portuguese; get in touch and we will find the best way to help.

chat Message us on WhatsApp
chat

Conta Walden4

Entrar

Use o e-mail e a senha da sua conta.